Group III selenides: Controlling dimensionality, structure, and properties through defects and heteroepitaxial growth

Semiconductors based on compounds of group III and group VI elements are receiving increased attention worldwide. These elements are found as either constituents or dopants in more traditional semiconducting materials such as group IV (Se, Ge), III–V (GaAs, AlAs, etc.), or II–VI (ZnSe, CdTe, etc.) materials. The III–VI compounds are chemically compatible with more traditional semiconducting materials, and they exhibit similar optical bandgaps and lattice constants, as shown in Fig. 1.^1,2 On the other hand, III–VI compounds exhibit many distinct structural, electronic, and optical properties that are not found in the traditional group IV, III–V, or II–VI materials. This combination of features holds significant promise for the development of semiconducting III–VI compounds as new device materials once issues related to heteroepitaxial growth on silicon and gallium arsenide substrates can be mastered. Layered III–VI materials also hold promise for isolated two-dimensional (2D) device applications.

The III–VI semiconductors comprise two principal stoichiometries: (a) $AIIIBVI$ compounds, such as GaS, GaSe, and InSe, and (b) $A2IIIB3VI$ compounds such as $Ga2S3$⁠, $Ga2Se3$⁠, and $In2Se3$⁠. In addition, $A2IIIBVI$ compounds such as $Ga2S$ exist as molecules and may be possible from the phase diagram; GaSe evaporates as $2GaSe→Ga2Se+1/2Se2$⁠.³ $A2IIIB3VI$ compounds exhibit a wide variety of crystal structures: the Crystallography Open Database² lists over two dozen distinct structures and polytypes for $AIIIBVI$ and $A2IIIB3VI$ crystals where A = (Al, Ga, In) and B = (S, Se, Te). Selected crystal structures for $AIIIBVI$ and $A2IIIB3VI$ semiconductors are shown in Fig. 2.

Bonding states built from s and p valence states have room for eight electrons per two atoms; III–VI atom pairs have nine valence electrons. This imbalance in electron counting for III–VI materials leads to bulk crystals containing atomic-scale voids lined with doubly occupied lone-pair orbitals. Group III-Se semiconductors crystallize in a variety of structures that incorporate neutral intrinsic vacancies relative to a parent zinc blende or wurtzite structure, where the vacancies may form lines [Figs. 2(b)–2(d)], planes [Figs. 2(a) and 2(f)], or be distributed throughout the crystal [Fig. 2(e)]. This leads to the orbitals lining these voids forming 1D, 2D, and 3D electronic structures.

$AIIIBVI$ compounds typically present a layered structure, exhibiting weak van der Waals bonding between separate, covalently bonded B–A–A–B layers [A = Ga, In; B = S, Se, Te; Fig. 2(a)]. These materials are often referred to as “van der Waals materials” and the ability to isolate single B–A–A–B layers holds promise for the development of two-dimensional electronics.⁴ Layered $AIIIBVI$ bulk crystals exhibit strong optical and electrical anisotropy^5–7 and high nonlinear optical coefficients in the infrared ranges, making them candidates for second harmonic generation materials.^8–13 This interest led to a vital activity in the 1970s and 1980s to fabricate bulk single crystals of GaSe and InSe, and many optical and electrical properties were investigated.¹⁴ 

Compounds of stoichiometry $A2IIIB3VI$ are more common, since the valence for group III and VI elements are +3 and −2, respectively. Chalcogenide-based $A2IIIB3VI$ materials, however, are considerably different from oxides such as $Al2O3$ and $B2O3$⁠. The lower ionicity of chalcogenides compared to oxygen leads them to prefer tetrahedral over octahedral bonding. They form a defected zinc blende or wurtzite structure, with one-third of the cation sites empty; for example, $Ga2Se3$ can be thought of as zinc blende ZnSe with every three $ZnII$ atoms replaced by two $GaIII$ atoms plus a vacancy. Different orderings of these empty sites along lines, helices, and planes lead to a variety of crystal structures (Fig. 2). These intrinsic vacancies, or atom-sized voids, allow $A2IIIB3VI$ to maintain semiconducting properties to a high level of impurity concentration¹⁵ and, in the case of Cr-doped $Ga2Se3$⁠, generate room temperature ferromagnetism.¹⁶ The crystal structure of these materials also explains their high radiation stability.¹⁷ Lithium atoms can fill these intrinsic voids in $(InxGa1−x)2Se3$ nanowires to generate new types of vacancy/lithium-atom ordered superlattices, demonstrating $A2IIIB3VI$ materials’ use in lithium ion storage, photovoltaics, and phase change memory.¹⁸ This is similar in concept to Li intercalation of a layered material, but Li resides on a helix instead of a plane. The intrinsic local electric field in layered $In2B3VI$ makes it a potential photocatalyst for splitting water.¹⁹ 

Initial interest in $A2IIIB3VI$ materials, in particular $Ga2Se3$⁠, stemmed from the interfacial reaction inherent to II–VI on III–V heteroepitaxy, where interface disruption caused by $Ga2Se3$ formation during the initial stages of ZnSe/GaAs interface formation was found to be a severe obstacle.^20,21 However, $Ga2Se3$ is a potentially useful material in its own right, especially in the thin film form. $Ga2Se3$ can passivate GaAs(001)²² and has strong potential for use in optoelectronic devices that exploit its visible bandgap.²³ When the vacancies are ordered, $Ga2Se3$ has large absorption anisotropy $(Δα>104cm−1)$⁠.²⁴ The close lattice matching of zinc blende $Ga2Se3$ to Si, GaP, and ZnS substrates (see Fig. 1) promotes its use in heterostructures. In particular, the close lattice matching (0.1%) and minimal interdiffusion (see below) at a properly formed $Ga2Se3$/Si interface suggests its use in silicon-based electronics, where the bandgap in the epitaxial films²⁵ lies in a convenient range for visible optoelectronics. The use of n-doped alloy $Ga2(S,Se)3$ deposited on p-Si heterojunctions for solar cells has been proposed.²⁶ 

Heteroepitaxy of III–VI materials is essential for incorporation of these novel materials into complex device structures. These materials also serve as vital prototypes addressing the basic physical chemistry governing heteroepitaxy of dissimilar materials. This Review addresses the structure and thin film growth of $AIII$-Se compound semiconductors using molecular beam epitaxy, primarily on silicon substrates, as well as the nature of bonding in bulk III–VI materials as it relates to controlling heteroepitaxy and the incorporation of transition metals for potential magnetic applications. The emphasis is on synthesizing work performed in collaboration between the Department of Physics and the Department of Materials Science and Engineering at the University of Washington. We both summarize work published in the archival literature and present work that has previously only been published in the dissertations of our graduate students.

In layered $AIIIBVI$ compounds, each structurally identical layer is composed of single planes of group VI atoms on either side of a double plane of group III atoms [Fig. 2(a)]. The basic building block is formed by stacking four hexagonal monoatomic sheets in a B–A–A–B sequence with trigonal prismatic symmetry. Different stacking and/or rotation by 180° of these covalently bonded layers into a superlattice results in several polytypes, only some of which exhibit inversion symmetry. Polarization-dependent linear and nonlinear optical properties of layered $AIIIBVI$⁠, therefore, depend on the polytype.²⁷ 

The B–A–A–B sequence of the quasi-two-dimensional $AIIIBVI$ layered semiconductors results in an unusual combination of chemical bonding types and hence electronic states.²⁸ Interlayer bonding is via weak van der Waals interactions, with strong intralayer interactions. The cation–cation bond is symmetric (covalent), while the cation–anion bond has a partial ionic character, as is evident from the total valence charge densities calculated using empirical pseudopotentials by Depeursinge for GaS, GaSe, and InSe.²⁹ In all three materials, valence electrons are strongly localized around the $BVI$ atoms, and the two $AIII$ atoms are linked by an $s$-like bonding charge. Increasing the ionicity from GaSe (fractional ionicity of 0.66) to GaS (0.74) and InSe (0.80) increases the ionic character of the $AIII–BVI$ bond, pulling the electronic charges more toward the $BVI$ atoms. In addition, when Ga is replaced with In or Se is replaced with S, the atomic $s$-states of the $AIII$ atom become less separated from the $BVI$ $p$-states, leading to delocalization of $AIII–AIII$ bond charge and weakening of the central bond (increasing the bond length). Extending this further results in InS not forming a layered structure, but rather a three-dimensional orthorhombic structure, because the atomic In 5s-states are at higher energy than the S 3p-states, changing the charge transfer.

The electronic states within the B–A–A–B layer as measured with angle-resolved photoemission for GaSe are well described with a simple linear combination of atomic orbitals.³⁰ Adjacent layers are weakly coupled, so the main difference between a single B–A–A–B quadlayer and the bulk crystal is a small splitting of the Se lone-pair states. Tight binding calculations show the $A–A$ bond is primarily of cation s and $pz$ character, while the three $A–B$ bonds per A or B atom primarily involve $pxy$ states (four electrons per anion, two per cation); the remaining anion electrons reside in doubly occupied lone-pair $s+pz$ orbitals on each surface of the layer. In GaSe, the hybridization of the Se orbitals is between that of $s+p3$ and $sp3$⁠: in pure $sp3$ bonding, all bond angles would be 109°, while for pure $s+p3$⁠, the intra- and bilayer angles would be 90° and 126°; in layered GaSe, the $Se–Ga–Se$ bond angle is about 100° and that between the central $Ga–Ga$ bond and any $Ga–Se$ bond is about 118°.³¹ As discussed below, both the electronic³⁰ and atomic^32,33 structure of one half of the quadlayer stays essentially the same when the central $Ga–Ga$ bond is replaced by an $Si–Ga$ bond during heteroepitaxial growth.

Semiconductors of the type $A2IIIB3VI$ typically crystallize in a structure based on zinc blende or wurtzite, in which one-third of the cation states are vacant, although layered or spinel structures are also reported. The structural vacancies order in different ways—in lines, helices, or planes—depending on both growth conditions and the choice of A or B atom. The use of the term “vacancy” is relative to the more familiar $AIIIBV$ or $AIIBVI$ zinc blende or wurtzite structures; it refers to neutral, atom-sized voids in the monoclinic $A2IIIB3VI$ unit cell. We also use the standard notation of the parent zinc blende and wurtzite crystals when referring to crystal surfaces and directions.

The bulk crystal structures derived from x-ray diffraction for zinc blende-based $β-Ga2Se3$ (Refs. 34 and 35) and wurtzite-based $Al2Se3$ (Ref. 36) are shown in Figs. 2(b) and 2(d), respectively. The structures differ by the [111] or [0001] stacking pattern of the $3×3$ arrays of vacancy lines, which are parallel to $[112¯]$ or $[112¯0]$⁠. In $γ-In2Se3$⁠, the defected wurtzite structure has vacancies aligned along a helix with an axis parallel to [0001], as shown in Fig 2(e).³⁷ In thin film growth on Si(001), vacancies in $Ga2Se3$ are aligned along $[11¯0]$ lines [Fig. 2(c)].³⁸ In these structures with linear or helical vacancy alignments, each cation (A) is bonded to four B atoms, while one-third of the anions have two cation neighbors and two-thirds have three. Under some conditions, $A2IIIB3VI$ forms B–A–B–A–B layers, with van der Waals bonding between them, as seen in $β-In2Se3$⁠, where the layers have hexagonal symmetry [Fig. 2(e)],³⁹ or in epitaxial $Ga2Se3$ on GaAs(001), where the cubic layer stacking is in the [001] direction (not shown).⁴⁰ The planar structure in Fig. 2(e) is similar to that of the topological insulator $Bi2Se3$⁠;$Bi2Se3$/$In2Se3$ superlattices are of interest to explore quantum size effects in topological insulators.⁴¹ The intrinsic vacancies in $A2IIIB3VI$ can lead to passivated surfaces (Sec. III B), since termination does not result in partially occupied, “dangling-bond” surface states, as well as extensive local structural rearrangements around impurities without broken bonds (Sec. VI B).

Potential commercial application of chalcogenide III–VI semiconductors requires heteroepitaxial thin films, since the intrinsic vacancy structure of bulk crystals generally leads to poor mechanical and thermal properties. Growth of these polar materials on a nonpolar substrate (such as silicon) requires understanding of the electron counting at the interface and any modification or interdiffusion required to reduce interface dipoles⁴² and/or passivate the surface dangling bonds.⁴³ Two important contributions related to electron counting are (a) the number of electrons available per atomic orbital and (b) the electrostatic potential difference generated when electrons transfer between atoms of different valence to fill those orbitals. Elimination of partially filled orbitals when bonds are broken at a semiconductor surface is a major driver of surface reconstruction; reduction of local electric fields at an interface between atoms of different valence is a major driver of interface mixing.

When III–VI materials, with nine electrons per atom pair, form a heterointerface with tetrahedrally bonded group IV, III–V, or II–VI materials, with eight electrons per pair, the imbalance in electron counting can lead to surface passivation (by fully occupying dangling-bond orbitals) and/or interface mixing. As described above, layered $AIIIBVI$ materials are of the form B–A–A–B, which may be thought of as two stable A–B bilayers that each have nine electrons per repeat unit [box in Fig. 3(d)]. One electron from each bilayer combines with one from the other bilayer to form the central A–A bond, six electrons from each bilayer form three A–B bonds, and the remaining two electrons fully occupy a “lone-pair” orbital on the layer surface [Fig. 3(d)]. This stable nine-electron bilayer with a single bond to another entity is ideal for passivating the surface of a tetrahedrally bonded semiconductor with eight electrons per unit cell when it is terminated in the plane perpendicular to a bond, i.e., the (111) surface of diamond or zinc blende or the (0001) surface of wurtzite.

In bulk silicon, each atom is tetrahedrally bonded along the $111$ directions. The surface created by cutting perpendicular to a bond direction thus has a singly occupied “dangling bond” for each surface atom. To lower its energy, the Si(111) surface reconstructs to a complex structure with a $7×7$ unit cell involving a combination of dimer formation, adatom adsorption, and stacking faults that result in only 19 dangling bonds instead of 49 per unit cell.⁴⁴ This complex structure is four layers thick and complicates heteroepitaxial growth on the Si(111) surface. It is possible, however, to remove the driving force for reconstruction by adding an electron to the dangling bond orbital. This may be accomplished, for example, by terminating the surface with hydrogen [Fig. 3(a)]. Simply exposing the Si$(111)7×7$ surface to hydrogen does not remove the reconstruction, but preparing an Si(111) surface chemically in buffered HF leads to a nonreconstructed, H-terminated $1×1$ surface where each valence electron of a surface Si atom is in a covalent bond (three with Si, one with H).⁴⁵ Note that the surface bilayer $(Si2H)$ has nine valence electrons (four from each Si, one from H) per surface unit cell and has a single bond perpendicular to the surface to the bulk crystal below.

An alternate method for passivating the Si(111) substrate and removing the reconstruction is to replace the top layer Si atoms with As, which has one additional valence electron, by cooling an Si surface in an As flux from above 800 °C.^46,47 The $Si–As$ bilayer also has nine valence electrons per $1×1$ surface unit cell, four from Si and five from As [Fig. 3(b)]; three of the As valence electrons bond with the underlying Si, and the remaining two occupy a “lone-pair” orbital at the surface. The Si(111):As $1×1$ surface is extremely stable, with no permanent changes after atmospheric exposure;⁴⁶ its formation during GaAs heteroepitaxy on Si(111) precludes laminar growth of GaAs, since its low surface energy relative to GaAs results in GaAs island formation.⁴⁸ 

A thought experiment that moves a proton from Si to As at the Si(111):As surface creates an AlSe bilayer with a very similar electronic structure except for the $Al–Se$ dipole. GaSe termination of Si(111) has the same electron counting [Fig. 3(c)]. Exposure of Si(111) to a flux of either gallium selenide^32,33 or aluminum selenide⁴⁹ leads to such a bilayer termination of the Si(111) surface, despite the absence of a stable layered crystal structure for bulk AlSe. The stable, unreconstructed Si(111):GaSe structure is extremely interesting in its own right and is discussed in Sec. IV.

The atomic structure of bilayer-terminated Si(111) is very similar to half of a bulk GaSe quadlayer.³⁰ Figure 4 shows the photoelectron diffraction results from Si(111):GaSe (Refs. 32, 33, and 50) and Si(111):AlSe.⁴⁹ Photoelectron diffraction yields element-specific structure through scattering of photoemitted electrons by neighboring atoms and is well modeled by multiple-scattering calculations,⁵¹ enabling extraction of structural parameters. At high kinetic energies [Fig. 4(a)], the largest feature comes from forward scattering of Ga emission by the surface Se atoms; Se emission shows only weak first order diffraction peaks from in-plane Se–Se scattering.^32,49 The dashed line in Fig. 4(a) highlights the 63° (65°) angle between the vertical and the Al–Se (Ga–Se) bond, which is close to the bulk GaSe value of 62°. At lower kinetic energies (KE),³³ electrons experience significant backscattering; a stereographic projection of emission from Ga $3d$ [Fig. 4(c)] shows peaks due to both the forward scattering from three neighboring Se at 65° and backscattering at normal emission from the underlying Si. The location of the interface Ga directly above the interface Si (top site), as opposed to three threefold hollows above the second (T4) or fourth (H3) layer Si, as well as the $Ga–Si$ bond length, were determined by comparing multiple-scattering calculations to the interference pattern of Ga $3d$ emission along the surface normal as the photon energy, and hence the electron wavevector, is varied [Fig. 4(e)]. Emission from Se atoms shows both backscattering from the three underlying Ga atoms at 65° and diffraction rings from both Ga and the six in-plane Se next-nearest-neighbors [Fig. 4(d)]. Varying the photon energy for Se $3d$emission along the Ga–Se bond [Fig. 4(f)] yields that bond length as well. These structural results as well as those from similar measurements on Si(111):AlSe (Ref. 49) are summarized in Fig. 4(b).

The similarity in atomic structure among Si(111):As, Si(111):GaSe, Si(111):AlSe, and GaSe (Fig. 3) continues to their electronic structure. Each exhibits a fully coordinated surface with a doubly occupied lone-pair $(s/pz)$ state at or near the valence band maximum that disperses downward from the center of the Brillouin zone, as measured with angle-resolved photoemission spectroscopy.³⁰ The states for a single bilayer of GaSe on Si(111) show a strong similarity to that of bulk GaSe, with the $Si–Ga$ bond states having a similar character to the $Ga–Ga$ states, and the Se lone-pair orbitals roughly the average of the bulk GaSe states that are split due to layer–layer interactions. The near-surface bands in Si(111):As, on the other hand, more closely resemble those of bulk Si.⁴⁷ The GaSe bilayer alters the Si bands for several atomic layers, reflecting strain, surface dipoles, or some other long-range effect. However, core-level photoemission shows the interface Si bonded to Ga to be in a bulklike environment. No interface shift (<0.1 eV) of Si $2p$ was resolved, unlike the 0.75 eV shift of the Si(111):As system. This serves as a reminder of the complex combination of initial and final state effects controlling core-level energies near a surface, as well as how surface band dispersion and charge transfer are separate issues. The similarity in the lone-pair dispersions between the bilayer and bulk crystal highlight the two-dimensional nature of the electronic states in bulk layered GaSe.

Despite the strong similarity in atomic and electronic structures of Si(111):AlSe and Si(111):GaSe, their resistance to contamination is quite different. As shown in Fig. 5, GaSe-terminated Si is much less reactive than Si(111):AlSe.⁵² When exposed to atmospheric-pressure air, nitrogen, oxygen, or nitrogen gas saturated with water vapor, Si(111):GaSe shows no change in surface-sensitive core-level photoemission [see left side of Figs. 5(b)–5(e) for result with atmospheric exposure]. Valence band emission [Fig. 5(a)] is suppressed by physisorbed species after atmospheric exposure (dashed line), but a 1 min vacuum anneal at 480 °C restores the system to the initial conditions [dotted line (green online)], including the Se lone-pair surface state near $−0.8eV$ (arrow). On the other hand, Si(111):AlSe permanently reacts with $O2$⁠, water vapor, or atmosphere at room temperature, exchanging O for Se. This is evident in the large (70%) decrease in Se $3d$ intensity [Fig. 5(c)], large O $1s$ emission [Fig. 5(e)], and the addition of a high binding energy component on Al $2p$ due to the higher electronegativity of oxygen relative to selenium [Fig. 5(b)]. The valence band structure permanently changes to open a wide bandgap upon exposure [Fig. 5(a)]. The oxidized AlSe bilayer does, however, protect the underlying silicon from reacting with the atmosphere as well as a GaSe bilayer does [Fig. 5(d)].

In contrast to AlSe and GaSe, InSe does not form a stable bilayer on Si(111),⁵³ despite the existence of a bulk layered InSe structure similar to that in Fig. 2(a). While exposure of Si(111) to evaporated GaSe or $Al2Se3$ results in an initial 1:1 ratio of cation to Se, even if additional Se or Al flux is supplied, respectively, InSe evaporation results in approximately one-third monolayer of Se deposited before In starts to stick. Subsequent growth is of $γ-In2Se3$⁠, as discussed in Sec. V C.

Creating a (001) surface of silicon breaks two bonds per surface atom, and the native surface reconstructs to form dimers in the $[110]$ or $[11¯0]$ direction (alternating with each layer), reducing the number of dangling bonds by half.⁵⁴ The Si$(001)2×1$ reconstructed surface is highly reactive, and when exposed to an Se flux [as in growth of ZnSe (Ref. 55) or $Ga2Se3$ (Ref. 50)], it forms SiSe_x similar to the Si$Ox$ formed when it is exposed to oxygen.⁵⁶ This reaction can be prevented by terminating the surface with arsenic. With five valence electrons, As forms dimers similar to the Si(001) surface, but instead of dangling bonds, each surface As exhibits a doubly occupied lone-pair orbital.⁵⁷ This passivated Si(001):As$2×1$ surface, which leads to islanded growth of GaAs on Si(001),⁴⁸ enables laminar heteroepitaxial growth of group III selenides, as discussed in Sec. VI A. Exchanging the outermost Si–As layer to create a stable Ga–Se bilayer on Si(001) in a manner similar to that discussed above for Si(111) has not been observed. A few bilayers of $Ga2Se3$ deposited on Si(001):As, however, form a stable structure resistant to oxidation.

The formation of $SiO2$ at the surface of Si(001) is the foundation of microelectronics. However, as devices shrink, a higher dielectric constant insulator is required to maintain a reasonable capacitance. Titanium dioxide, which in its anatase crystal structure is nearly lattice matched to silicon, is a candidate high dielectric constant material. However, its exploitation is hindered by the formation of a low-dielectric-constant interface $SiO2$ layer that dominates the system capacitance.⁵⁸ This problem can be remedied through the use of surface passivation, namely, Si(001):As:$Ga2Se3$⁠.⁵⁹ In contrast to clean Si, Si(001):As is stable when exposed to $O2$⁠, even at 450 °C; the black line in Fig. 6 shows a silicon core level (b) and valence band emission (d) identical to clean Si(001):As. When reactive Ti is added to the oxygen, however, substrate oxidation occurs for Si(001):As, with reacted components present in both the Si $2p$ [Fig. 6(b), dots (blue online)] and As $3d$ (not shown) emission. On the other hand, a 0.6 nm buffer layer of $Ga2Se3$ on top of Si(001):As prevents oxidative disruption of the interface and enables laminar growth of $TiO2$ without the formation of $SiOx$ [note the absence of oxide peak for Si $2p$ in Fig. 6(a)].⁵⁹ The valence band offset between Si and $TiO2$ is similar both with [Fig. 6(c)] and without [Fig. 6(d)] the $Ga2Se3$ buffer layer, with the Fermi level pinned near the top of the oxide bandgap [solid lines (red online)] and the bottom of the silicon gap (dotted lines). The band alignment was deduced from knowing the bulk separation between the valence band maxima and the $2p$ levels for Ti and Si, as well as the bulk bandgaps.⁵⁹ Unfortunately, this staggered type-II band alignment is not appropriate for using anatase as a gate insulator for n-type silicon. The structure, growth, and properties of this $Ga2Se3$ buffer layer are detailed in Sec. VI.

The equivalent surface of GaAs to Si(111) can be either Ga-terminated $(111)$ or As-terminated $(1¯1¯1¯)$⁠. These polar surfaces have an intrinsic dipole as well as dangling bonds, leading to a complex reconstruction that is different on the two surfaces. On GaAs $(111)$⁠, removing 1/4 of the surface Ga leads to a $2×2$ reconstruction with no partially filled orbitals; GaAs $(1¯1¯1¯)$ takes either a $2×2$ or $19×19$ reconstruction, depending on the As concentration.⁶⁰ 

Electron counting arguments for GaAs surfaces are more complicated than those for silicon since the GaAs bond is not symmetric: in the bulk, each Ga–As bond averages $3/4$ of an electron from Ga and $5/4$ of an electron from As. At the ideal $(1¯1¯1¯)$ surface, there is thus $5/4$electron per As dangling-bond orbital; to fill these orbitals, three electrons thus need to be added for every four surface anions. This may be accomplished by replacing three out of four $AsV$ atoms with $SeVI$⁠. Combined photoemission and RHEED of GaSe deposition on GaAs$(1¯1¯1¯)$ is consistent with this picture, with a stable $GaAs0.25Se0.75$ bilayer bound to the substrate, followed by subsequent deposition of stoichiometric, layered GaSe (see Fig. 7, right).⁶¹ At the ideal (111) surface of GaAs, the Ga-terminated surface will have its dangling bond orbitals each occupied by $3/4$ electron. Removing one $GaIII$ atom out of four thus leads to a surface where each Ga–As bond has two electrons and the Ga-derived dangling bond orbitals are all empty. Photoemission spectroscopy⁶¹ shows that the deposition of GaSe on GaAs(111) results in extensive interdiffusion of As and Se, and ∼1 ML of Se is observed as an extra layer between the GaAs substrate and subsequent deposition of layered GaSe (Fig. 7, left). Figure 7(c) compares the GaSe/GaAs interface for the two surface terminations; note the different location and bonding of the Se interface layer as well as the enhanced interdiffusion on the Ga-terminated (111) surface. In both cases, the Se-passivation leads to a discommensurate interface between the GaSe and lattice-mismatched underlying substrate.⁶¹ 

GaSe-terminated Si(111) is an extremely stable, unreconstructed surface that leaves the underlying Si in an essentially bulklike environment. This quasi-two-dimensional material is important in its own right as a prototype system to study the formation of domain boundaries (DBs), charge redistribution near substitutional defects, and spin–orbit effects in photoelectron diffraction.

As discussed in Sec. III, exposure of the reconstructed Si$(111)7×7$ surface to evaporated GaSe completely removes the complex reconstruction. Figure 8 contains scanning tunneling microscopy (STM) images of Si(111) with [Figs. 8(b) and 8(d)] and without [Figs. 8(a) and 8(c))] a terminating GaSe bilayer.^62,63 On the nanometer scale [Figs. 8(c) and 8(d)], the addition of GaSe reduces the surface corrugation from hundreds to tens of picometers as adatoms and deep corner holes are eliminated (⁠$7×7$ unit cell denoted by the gray (green online) rhombus is 2.7 nm on a side). The Si(111):GaSe images are very similar to STM of Si(111):As.⁶⁴ Atomic corrugations in GaSe bilayer-terminated Si(111) can be visualized with both positive (empty state) and negative (occupied state) bias STM, with the peaks in tunnel current at the same location (above Se atoms) in each case.⁶² The Si(111):GaSe island, pit, and corrugated step-edge structure apparent in Fig. 8(b) results from the displacement of Si atoms when the $7×7$ structure, which is four layers deep and has four more atoms per unit cell than the ideally terminated surface, reacts to form the bilayer-terminated surface.

Three types of intrinsic defects with different dimensionalities are observed with in situ STM for GaSe bilayer-terminated Si(111): zero-dimensional clustered point defects (CPDs), one-dimensional DB, and two-dimensional Ga-terminated Si regions (Ga/Si).⁶² DBs are formed during the coalescence of two GaSe bilayer domains whose orientations differ by 180°: one with Ga–Se bonds in same direction as Si–Si bonds (zinc blende or diamond stacking), and hence Se in the H3 (hollow) site, and the other with Ga–Se bonds antiparallel to Si–Si bonds (wurtzite or layered GaSe stacking), and hence Se in the T4 site directly above the second-layer Si. The rotated (wurtzite-orientation) domains are kinetically stable only when the films are grown at temperatures below 550 °C (which is the growth temperature for the single-domain GaSe films in Fig. 4); the local orientation is likely related to whether nucleation occurs on the faulted (GaSe stacking) or unfaulted (Si-stacking) half of the $7×7$ unit cell. These orientational domains are distinguished by an apparent height difference of ∼0.5 Å in the empty state imaging by STM [Fig. 9(a)] but have similar apparent heights in occupied states [Fig. 9(b)]. A model deduced from high resolution imaging of the empty [Fig. 9(c)] and occupied [Fig. 9(d)] states at a DB includes a Ga dimer structure similar to the Si dimers found at the boundary between faulted and unfaulted regions of the Si(111) $7×7$ unit cell, as well as occasional pairwise substitution of Se atoms by Si at the domain boundary that likely reduces the electrostatic energy of the Ga–Se dipole interaction across the boundary.⁶² 

Orientational domain boundaries frequently terminate in Ga-terminated Si regions, where Ga segregates under Se deficient conditions. The dark region just below the inset in Fig. 9(b) is an example. A larger Ga-terminated region is imaged in Fig. 9(e). The structure is similar to the $6.3×6.3$ structure reported for about 1 ML of Ga on Si(111).⁶⁵ These domains, whose existence may also be detected through a low-binding energy component in Ga $3d$ photoemission, are suppressed by including an extra source of Se during growth.⁵⁰ 

Isolated point defects are not observed with STM for the GaSe bilayer; rather, defects appear as CPDs involving several unit cells. For example, the inset in Fig. 9(a) shows three 0.4 Å protrusions at Ga positions separated by three unit cells; these appear as 0.1 Å dents in occupied state imaging [inset in Fig. 9(b)]. Ohta et al.⁶² reported several distinct CPDs, all of which extend over at least three unit cells and propose that these CPDs are best explained by a localized charge density wave that scatters from a substitutional point defect and are evidence of intermixing or interdiffusion during growth. The CPD structures observed with STM differ for the two domain orientations. The CPD in the insets in Figs. 9(a) and 9(b) is the most common defect for the rotated domains; it is centered on the location of a Ga atom and is attributed to substitutional $SiGa$⁠. The most common defect on nonrotated domains appears as a protrusion centered on an Se site with an additional depleted hexagon (sides of length two unit cells) in occupied state images that is not present in empty state images. Several may be seen in the upper-right of Figs. 9(a) and 9(b). In both of these cases, the STM image reflects nanometer-scale charge redistribution within the 2D bilayer in the presence of a 0D substitutional point defect.

The highly ordered structure of Si(111):GaSe, with its shallow Ga $3d$ core level (binding energy $∼20eV$⁠), provides a fruitful testing ground to investigate the role of spin–orbit interactions in photoelectron diffraction. In the case of As-terminated silicon, differences in the angular variation of intensity for As $3d5/2$ and $3d3/2$ emission were attributed solely to the difference in energy of the two states,^66,67 while for In-terminated Si(100), additional structure in the In $4d$ spin–orbit branching ratio (SOBR, ratio of $d3/2$ to $d5/2$ emission intensities) was attributed to unknown matrix element effects.⁶⁸ Si(111):GaSe displays a surprisingly strong (>15%) angular variation of the Ga 3d photoemission SOBR.^50,69 Figure 10(a) shows the angular variation along the $[112¯]$ azimuth for the full Ga $3d$ emission [equivalent to horizontal diameter in Fig. 4(c)] and separately for the spin–orbit components, $3d5/2$ and $3d3/2$ $(hν=246eV)$⁠. The sample was rotated while the angle between the linearly polarized incident photons and detected electrons was held at 60° in the plane of the photon polarization. The inset in Fig. 10(a) shows the distinct variation in the Ga $3d$ peak shape as the spin–orbit ratio varies with the emission angle; the angular variation of the SOBR is shown in Fig. 10(b).

The origin of the variation in SOBR with the angle can be addressed through calculation by adapting relativistic photoabsorption matrix elements, originally developed for modeling x-ray absorption fine structures,⁷⁰ and applying them to multiple-scattering analysis of photoelectron diffraction.^50,69 Figure 10(c) shows results of such calculations using FEFF,⁷⁰ which features an ab initio treatment of atomic properties based on a spinor-relativistic single-configuration Dirac–Fock atom code. The SOBR for Ga $3d$emission from Si(111):GaSe is shown both with (solid line) and without (dotted line) taking into account the 0.4 eV difference in their kinetic energies, assuming the structure of Fig. 4(b), as well as that ignoring scattering from neighboring atoms (dashed line). The results show that SOBR angular variation is not strongly dependent on the energy difference between the two spin–orbit components nor is it an atomic feature. Rather, the SOBR angular variation is the result of different diffraction conditions for the two spin–orbit components due to different final state wave functions; qualitatively, it arises from their different $p-$ and$f-$wave decomposition. The calculations predict both the position and relative amplitude of the Ga SOBR variations. The absolute amplitude of the experimental effect is roughly four times larger than the theoretical prediction for Ga $3d$ emission [compare Figs. 10(b) and 10(c)]. On the other hand, the variation of the $Se3d$ SOBR (not shown) is about $±3%$ for both experiment and theory. The experimental enhancement of the Ga $3d$ SOBR variation may be attributed to overlap between Ga $3d$ and Se $4s$ levels; these levels were not included in the theoretical calculation but would increase the overlap between the initial and final states for Ga. Another factor impacting both Ga and Se modeling is the use of spherical muffin-tin potentials for the calculation despite the presence of a crystal field, which may account for the higher average branching ratio in the calculation than in the data for both Ga and Se.⁵⁰ 

We turn now to the initial stages of heteroepitaxial growth of $AIIISe$ semiconductors on silicon. The wide variety of stable structures for III–VI materials means that the choice of substrate termination and orientation can determine the structure and stoichiometry of the III-Se film, and hence its potential uses. All the structures in Fig. 2 involve different stackings of hexagonal layers with threefold symmetry, which is the structure of the Si(111) surface. The (001) plane of silicon, on the other hand, has square atomic arrangements with two perpendicular mirror planes. Of the common $AxIIIByVI$ crystal structures, only $β-Ga2Se3$ (defected zinc blende) is cubic, although there is⁷¹ a high pressure defected-spinel phase for $Al2Se3$⁠; the other structures are mainly hexagonal. This raises the question of whether heteroepitaxial growth of group III selenides on Si(001) and Si(111) will lead to different structures or if weak van der Waals bonding can lead to discommensurate growth of hexagonal layered materials in either case. Also, the bulk cubic phase $β-Ga2Se3$ exhibits vacancy lines along the $[112¯]$ direction, which is perpendicular to [111], but not to [001]; this leads to different growth dynamics and vacancy ordering on the two surfaces. In this section, we discuss the nucleation and growth of aluminum, gallium, and indium selenide on Si(111); Sec. VI discusses aluminum and gallium selenide growth on Si(100).

Aluminum selenide first forms the Si(111):AlSe bilayer termination discussed in Sec. III, despite not having a bulk layered form similar to Fig. 2(a); it then continues growth along as defected wurtzite $Al2Se3$⁠, with $[0001]Al2Se3∥[111]Si$⁠.⁷² Gallium selenide, on the other hand, continues growth on the Si(111):GaSe bilayer termination as layered GaSe.⁷³ Indium selenide, which does exhibit a bulk InSe layered structure,⁷⁴ does not form bilayer termination; it first terminates the Si(111) surface with a partial monolayer of Se and then continues as $γ-In2Se3$⁠, although with a complex surface reconstruction.^53,75 The initial film morphology for deposition of $AIIISe$on Si(111), as imaged with scanning tunneling microscopy, is shown in Fig. 11 for evaporation source materials of $Al2Se3+Al$ (top), GaSe (middle), and $In2Se3$ (bottom).

Exposure of Si(111) to evaporated GaSe $(Ga2Se+0.5Se2)$ at substrate temperatures in the range 400–585 °C results in essentially stoichiometric GaSe deposition throughout, except for a small fraction of, the Ga-terminated regions [e.g., Fig. 9(e)] that may be eliminated by adding extra $Se2$ to the flux to counteract its smaller initial sticking coefficient $relativetoGa2Se$⁠. Initially [Fig. 11(d)], GaSe forms small islands that later merge to form a flat bilayer, followed by subsequent nucleation and growth of layered GaSe once the bilayer is complete.⁶³ Photoemission reveals the ratio of Ga to Se to remain constant (and equal to that of bulk GaSe) throughout this evolution. The sub-bilayer growth imaged in Fig. 11(d) shows small islands of height close to that of the smooth, completed bilayer visible along the left of the image; disordered $7×7$ structure is present between the islands. Figure 11(e) shows a film just after completion of the GaSe bilayer; the bright islands are 0.8 nm high, which is the height of the Se–Ga–Ga–Se quadlayer, easily distinguished from the 0.3 nm step height between bilayer-terminated regions [rendered as black and light gray (orange online)] that is characteristic of the silicon substrate.

Once the bilayer is complete, the GaSe growth rate decreases, with only about 10% of the atoms sticking above 450 °C; there are very few nucleation sites for the next layer. The main nucleation sites are at domain boundaries, with additional sites at substrate step-edges, as seen in the larger scale image in Fig. 12(a) and expanded view in Figs. 12(b) and 12(c). Figure 12(b) images a collection of triangular GaSe islands a few nm in size; this film was deposited at $Tsub=400°C$ so that the bilayer exhibits many domain boundaries, and shows that most islands inhabit one of two orientations that differ by 180°. Figure 12(c) is the horizontal gradient of the image in Fig. 12(b), highlighting orientational domain boundaries; most islands show a domain boundary terminating under them (circles). With further deposition, growth continues as layered GaSe, with its smooth surface and 0.8 nm steps. Initially, the layers confine themselves to a single substrate terrace [Fig. 12(d)], but once the second layer nucleates, the flexible layers eventually “carpet” the substrate steps [inset in Fig. 12(e)].⁶³ 

The structure of sub-bilayer aluminum selenide growth on Si(111) [Fig. 11(b)] shows similar structures to gallium selenide deposition [Fig. 11(d)], although AlSe tends to form lines along the $[11¯0]$ direction rather than the compact GaSe islands on the $7×7$ structure. In Fig. 11(b), the smooth triangle is a complete AlSe bilayer. Once the bilayer is complete, additional nucleation and growth is as defected wurtzite $Al2Se3$⁠. In Fig. 11(c), the step height of the islands atop the AlSe bilayer is 0.3 nm, equal to a single hexagonal bilayer (HBL) of the wurtzite structure. The islands exhibit a $3×3R30°$ structure consistent with 1/3 of the Al sites being vacant, this reconstruction continues as the surface as the film grows thicker.⁷² 

Surface-sensitive core-level spectroscopy, which probes interface chemistry and stoichiometry, is shown in Fig. 13 for (a) aluminum selenide⁷² and (b) indium selenide⁵³ growth. Films were deposited while slowly moving a shutter across the substrate to create a wedge-shaped sample, and photoemission spectra were acquired as a function of position, and hence film thickness. The photoemission peaks for aluminum selenide growth [Fig. 13(a)] show that the bilayer component of Al and Se emission (shaded, dashed line) is steadily buried at the same rate as emission from the underlying silicon, indicating that the bilayer remains intact. The single Al peak and two Se components separated by about 1 eV that arise with additional HBL have the same energies as those for an 8 nm-thick film,⁷⁶ showing that the bulk $Al2Se3$ structure depicted in Fig. 2(d) starts as soon as the first bilayer is completed. High energy photoelectron diffraction of these emission peaks exhibits the symmetry and expected positions for wurtzite.⁷⁶ 

While AlSe and GaSe form a bilayer termination of Si(111), InSe does not.⁵³ With a large lattice mismatch (see Fig. 1), indium does not readily fit into the Si structure. Instead of forming In–Si bonds, the growth initiates instead with Se replacing the adatoms in the Si(111) $7×7$ structure. Core-level spectroscopy [Fig. 13(b)] shows that a partial Se layer sticks to the surface before any indium, but the Se:In ratio quickly converges to 3:2 by the time one monolayer of In has been deposited. High resolution scanning tunneling microscopy for the Se-only layer [Fig. 11(f)] shows protrusions at the position of the Si adatoms; these protrusions appear taller in occupied state images than in empty state images, consistent with $SeSi$⁠.

Further deposition of indium selenide on Si(111) in the temperature range $Tsub$ = 570 –580 °C results in growth of defected wurtzite $γ-In2Se3$⁠.^53,75 Lower temperatures tend to result in disordered, Se-rich, rough films, while higher temperatures are In-rich with very little Se.⁷⁷ The incomplete HBL coverage imaged in Fig. 11(g) shows atoms scattered seemingly randomly and occasionally clustering to form ordered pinwheel shapes with a layer height of 1.4 Å above the base level; the uncovered regions (dark triangle) exhibit a local $3×3R30°$ reconstruction. This $3$ pattern is consistent with 1/3 ML of Se in adatom sites, while the atoms above these are likely indium. The 1.4 Å step height is consistent with adding In and then incorporating the underlying Se into a bilayer structure. Additional deposition leads to coalescence of the pinwheel shapes into hexagonal figure-eight shapes [Fig. 11(h)].

As more $In2Se3$ is added,^53,77 the layers maintain the same surface structure with steps of height 0.3 nm, as expected for defected wurtzite $γ-In2Se3$⁠, and not the 1.0 nm steps of layered $β-In2Se3$ or the 0.83 nm spacing of layered $InSe$⁠. Despite the large in-plane mismatch, $In2Se3$ does not form islands to relieve interfacial stress; rather, the hexagonal lattice constant of the $γ-In2Se3$ layer is very close to its bulk value starting at the first layer, resulting in a discommensurate interface. The surface forms a canted honeycomb pattern oriented at angles of about ±2° with the substrate $11¯0$ directions [Fig. 14(a)]. The Fourier transform of the STM image in Fig. 14(a) is shown in Fig. 14(b), forming a hexagonal pattern where each vertex is a pair of spots. Low energy electron diffraction (LEED) from this surface at 16.4 eV [Fig. 14(c)] shows the underlying hexagonal $1×1$ pattern plus superstructure spots consisting of two triangles; the triangles indicate a total of six domains that are nearly rhombohedral, but with lattice vectors of two slightly different lengths. Electron diffraction at 34.2 eV shows a more complicated pattern [Fig. 14(d)] with multiple superstructure components. The LEED data are well modeled⁷⁸ with six domains (three left-handed and three right-handed) at ±2° from the three symmetry-equivalent $11¯0$ directions on the surface, with lattice constants 1.72 and 1.84 times the underlying $γ-In2Se3$ lattice vector of 4.14 Å (⁠$7.14$ and $7.63Å$⁠).⁷⁷ 

STM of several-nm-thick films of $In2Se3$ deposited at 575 °C shows that the complex, discommensurate reconstruction persists at the surface, while x-ray diffraction shows that the underlying film exhibits the $γ-In2Se3$ defected wurtzite structure.⁵³ When amorphous $In2Se3$ is deposited on Si(111) at room temperature and then annealed to 380 °C, it crystallizes to $γ-In2Se3$ without the complex reconstruction; STM shows smooth terraces with step-height 0.3 nm and both LEED and STM exhibit a $3×3R30°$ surface reconstruction from ordered In vacancies. The crystallized material can be returned to the amorphous state with a high energy laser pulse, indicating potential application of $In2Se3/Si(111)$ films for phase change memory.⁷⁵ 

In summary, the evolution of nanostructure morphology and local chemical environment during heteroepitaxial growth of the group III (Al, Ga, and In)–selenides on Si(111) lead to a variety of structures and morphologies. Despite the strong similarity between AlSe and GaSe in atomic and electronic structure during the deposition of their first bilayers, subsequent growth is quite different. Once the bilayer is completed, nucleation and growth of Al_xSe_y results in an alternating Al–Se–Al–Se stacking sequence consistent with the defected wurtzite structure of Al₂Se₃, whereas gallium and selenium exhibit the layered GaSe structure, with van der Waals bonding between covalently bonded Se–Ga–Ga–Se layers. In the case of indium selenide, an InSe bilayer is not formed, but it first terminates the Si(111) surface with a partial monolayer of Se, and then continues as $γ-In2Se3$⁠, although with a complex surface reconstruction. Potential use of bilayer-terminated Si(111) for heteroepitaxy of other materials has not been fully explored, although its low surface energy makes it likely that materials which are not intrinsically layered will exhibit islanded rather than laminar growth, similar to growth⁴⁸ of GaAs on Si(111):As. Attempts to grow GaSe on As-terminated Si(111) resulted in about two bilayers of a cubic $Gay(AsxSe1−x)$ interface layer, but the growth then switched to layered GaSe.⁶³ 

The growth of group III selenides on Si(001) is complicated by two factors: the square symmetry and the high reactivity of the (001) surface with selenium. The former leads to novel crystal structures constrained by the substrate symmetry, while the latter can be overcome through the use of arsenic passivation. As discussed in Sec. III, the same tendency of Si(001) to react with oxygen to form amorphous $SiO2$ leads to a similar reaction with selenium, but passivation of the Si(001) surface via As-termination enables the growth of selenides without forming amorphous $SiSex$⁠.

Scanning tunneling microscopy and surface-sensitive core-level spectroscopy for aluminum⁵³ and gallium⁶³ selenide growth on Si(001):As are shown in Figs. 15 and 16, respectively. The Si(001):As surface consists of rows of As dimers along $11¯0$ directions that rotate by 90° with each layer [Fig. 15(a)]. The first selenide layer nucleates differently for $Ga2Se3$ and $Al2Se3$ deposition: $Ga2Se3$ nucleates in lines that are perpendicular to the As dimer rows immediately beneath it [Fig. 15(g)], while $Al2Se3$ nucleates parallel to the underlying dimer rows [Fig. 15(b)]. Photoemission spectroscopy of the underlying substrate shows that $Al2Se3$ deposition does not dramatically change the line shape of the Si or As emission, indicating the absence of strong bonding or interdiffusion [Figs. 16(c) and 16(d)], while for $Ga2Se3$ deposition, the interface silicon changes to a more bulklike environment (suppression of the high binding energy peak associated with Si bonded to two arsenic atoms) and the As $3d$ broadens to contain at least two components [Fig. 16(b)]. The narrow Si $2p$ interface emission is indicative of forming Si–Ga bonds at the interface, which, similar to the (111) case, exhibit no chemical shift from the bulk, instead of Si–As bonds, which exhibit a 0.5 eV chemical shift. The As spectrum [Fig. 16(c)], as well as its photoelectron diffraction pattern,⁶³ indicate substitutional $AsSe$ resulting from diffusion into the first couple of cubic bilayers in the overlayer.

As $Al2Se3$ growth proceeds, the morphology is most ordered after the deposition of one to two cubic bilayers (CBL). Narrow $(1−3)$ rods, several nm long, alternate direction with each substrate terrace step [Fig. 15(c) and 15(d)]; Si(001):As with dimer rows parallel to the nanorod structure is still visible between the nanorods, which accounts for the persistent photoemission intensity from the substrate [Fig. 16(d)] and As interface layer [Fig. 16(c)]. Se $3d$ emission [Fig. 16(b)] from the first one to two CBL exhibits a peak of similar half-width to that seen for (111) growth of $Al2Se3$⁠.⁷² Once the film is more than about two BL thick, however, the Se core-level loses structure and broadens, indicating a disordered film, and STM shows the nanorod structure to break up into smaller and smaller pieces that show much less alignment with the underlying substrate [Fig. 15(e)].⁵³ 

In contrast to $Al2Se3$⁠, the quality of $Ga2Se3$ heteroepitaxy on Si(001):As improves with increasing thickness. The first CBL coalesces into smooth, broad anisotropic islands with rough edges; the substrate dimer rows do not persist [Fig. 15(h)]. By about two CBL, elongated nanorods develop smooth edges and a common width of about 3 nm, while photoemission (Fig. 16, right) shows development of an interface dipole between the substrate Si [Fig. 16(d)] and the overlayer (overlayer peaks shift to higher binding energy faster than the Si); the overlayer contains single-component gallium [Fig. 16(a)] and dual-component selenium [Fig. 16(b)]. The As $3d$[Fig. 16(c)] integrated intensity decays with thickness more slowly than the Si $2p$ [Fig. 16(d)], while photoelectron diffraction shows the As to have the same diffraction pattern as Se (and different from Ga),⁶³ indicating As is substituting for Se in the $Ga2Se3$ structure and not all remaining at the interface. Once the film is three CBL thick, the nanorods sharpen to less than a nanometer wide. As growth continues beyond three CBL, the rods maintain this minimum spacing of three dimer rows (1.2 nm), and minimum height differences of 0.26 nm, the CBL spacing.^38,63

High resolution imaging of the stable $Ga2Se3$ nanorod surface [Fig. 15(k)] shows 10 pm atomic corrugation with periodicity 3.9 Å, the bulk (second-neighbor) distance between adjacent Ga or Se atoms.³⁸ Comparison of empty and full state images, as well as photoelectron diffraction and comparison of photoemission intensities at different photon energies (and hence escape depths), shows the surface to be Ga terminated.⁶³ These one-dimensional nanorods lead to the formation of one-dimensional vacancy rows in the film as the growth proceeds [Fig. 15(f)]. Angle-resolved photoemission shows a one-dimensional electronic state along [$11¯0$] near the valence band maximum that is evidence for vacancy alignment in this direction throughout the film;⁷⁹ this state is similar to calculations⁸⁰ of Se lone-pair states lining the $[112¯]$ vacancy rows in $β-Ga2Se3$⁠.

The striking difference between growth morphologies of $Al2Se3$ and $Ga2Se3$ on Si(001) is strongly connected to the behavior of arsenic at the interface.^53,63 Given that the nanorods are cation terminated, the initial alignment of the $Al2Se3$ nanorods parallel to the underlying As dimers requires that aluminum atoms are inserted into the As dimer bond for a zinc blende stacking arrangement [Fig. 17(a)]. At the level of one or two CBL, the resulting growth, with Al vacancies aligning in $11¯0$ directions, is stable. However, with further deposition, the field arising from the large dipole associated with the intact polar/nonpolar interface leads to instability, which, when combined with the tendency for $Al2Se3$ to take a wurtzite rather than zinc blende structure, results in the breakup of the nanorod structure. For $Ga2Se3$ growth, on the other hand, the exchange of As with Ga at the interface [Fig. 17(b)] leads to a nonpolar bond (Si $2p$ shows no chemical shift), and the electron counting resulting from a combination of Ga vacancies and arsenic interdiffusion allows the surface to create and maintain a structure with no net dipole and no dangling bonds. For the first two to three layers, As interdiffusion leads to fewer vacancies (⁠$3AsSe$have the same electron counting as one $VGa$⁠), which suppresses the nanoridge formation; once the As supply is depleted, the aligned nanoridge structure stabilizes. The stability of this structure and the absence of partially filled surface orbitals account for the extreme resistance of this structure to oxidation, even in the presence of titanium at elevated temperature, as discussed in Sec. III B.

Commercial application of semiconductors relies on doping with impurities. The intrinsic vacancies inherent in $A2IIIB3VI$ semiconductors raise the question of how dopants may incorporate in these crystals. For example, a group-II dopant atom could replace a group III atom, and thus act as an acceptor, or occupy a vacancy, and thus act as a double donor. Besides electronic considerations, intrinsic-vacancy materials raise interesting scientific questions related to the potential mechanisms for ferromagnetism in dilute magnetic semiconductors, namely, mediation by free carriers [as in Mn:GaAs (Refs. 81 and 82)] or stationary defects (as in $Co:TiO2$⁠).⁸³ In GaAs, Mn doping at the level of a few percent leads to low-temperature ferromagnetism, with Mn contributing both the localized spins (five unpaired $3d$ electrons) and the itinerant holes that couple them [two bonding electrons $(4s2)$ where Ga would have three (⁠$4s2p$⁠)].⁸¹ In Co-doped anatase $TiO2$⁠, ferromagnetism has been observed in insulating films, with the coupling between Co spins attributed to localized defect states.⁸³ With defect-associated states at the top of the valence band, namely, the Se lone-pairs that line vacancy rows,^79,80 $Ga2Se3$ has potential for both mechanisms to contribute. Density-functional calculations of manganese, chromium, or vanadium inserted into a vacancy site in $Ga2Se3$ suggest that any of these dopants may result in spin-polarized $3d$ states, with the largest net spin near the Fermi level predicted for Cr.^84,85 These calculations, which were for a structure that nature did not choose to make in our experiments, nevertheless inspired exploration of Mn and Cr doping at the level of a few atomic percent in epitaxial $Ga2Se3$ films for potential magnetic applications. As discussed below, epitaxial $Mn:Ga2Se3$ is not a promising dilute magnetic semiconductor, but epitaxial $Cr:Ga2Se3$ exhibits room temperature ferromagnetism. Both dopants, however, have significant impact on the heteroepitaxial growth of $Ga2Se3$⁠.

Mn doping of $Ga2Se3$ films on Si(001) leads to disproportionation into lattice-matched MnSe islands and a low-Mn-concentration laminar film before the doped film is either thick enough or at a high enough concentration of Mn to have significant magnetic response.^86,87 Chromium doping, on the other hand, leads to room temperature ferromagnetism.¹⁶ The unique vacancy structure of $Ga2Se3$ allows Cr to incorporate in a locally octahedral environment without major distortion of the overall lattice, while the Cr $3d$ states hybridize with the Se lone-pairs, resulting in laminar, ferromagnetic films.

Manganese is soluble in bulk $Ga2Se3$ up to $x=0.15$ in $Mn3xGa2(1−x)Se3$⁠, after which there is first precipitation of the quasi-zinc blende $x=0.25$compound $MnGa2Se4$⁠, and then, for $x≥0.25,$disproportionation into $MnGa2Se4$ and rock salt MnSe, although zinc blende-based $Mn3xGa2(1−x)Se3$ is observed to higher Mn concentrations above 930 K.⁸⁸ For growth on Si(001), however, both zinc blende $Ga2Se3$ and rock salt MnSe are better lattice-matched to the substrate than either $MnGa2Se4$ or the bulk substitutional alloy with $x>0.05$⁠. This leads to disproportionation into MnSe islands during attempts to fabricate Mn-doped $Ga2Se3$ films on Si(001).^86,87

The morphology of Mn-doped $Ga2Se3$ films deposited on Si(001):As varies with both thickness and Mn concentration, and to a lesser extent on the presence or absence of an undoped $Ga2Se3$ buffer layer (see Fig. 18).^86,87 Mn concentration in Fig. 18 was determined from the incident flux, measured as $z=MnQCM/(MnQCM+GaSeQCM)$⁠, where QCM denotes the room temperature quartz crystal monitoring the flux. At low Mn concentrations and/or thickness (e.g., sample D1 in Fig. 18, which was deposited on a 4° miscut wafer to align domains), films are laminar with a nanoridge structure similar to that described above for pure $Ga2Se3$ films [e.g., Fig. 15(j)]. As the amount of deposition and/or Mn concentration is increased, there is a transition to an islanded morphology, with flat rectangular islands protruding from terraces that exhibit a distribution of nanoridge spacings and orientations similar to pure $Ga2Se3$ films; the specific island morphology varies with the growth parameters (compare D7 and D8 in Fig. 18), but island edges are generally aligned with $[11¯0]$ and $[110]$ directions. An intermediate “platelet” regime (e.g., sample D5 in Fig. 18) has tall narrow islands nucleated on terraces that contain platelets of about 30 oriented and uniformly spaced $(d∼2.8nm)$ nanoridges. The platelet morphology is metastable, converting to the islanded structure upon a 30 min anneal at the growth temperature of 500 °C. Samples with an $∼1nm$ buffer layer of pure $Ga2Se3$ exhibit islands with a somewhat larger footprint (compare D8 and B4 in Fig. 18), but the local atomic structure of both the islands and the terraces between them are similar with and without a buffer layer.